If you are thinking about using Search Console's new setting to opt out of AI Overviews, the first public test of it is a warning. When SEO consultant Cyrus Shepard opted his own site out of Google's AI features for 28 days, he estimates the site lost about 21% of its Google search clicks, from results that made up only about 10% of its impressions.

Our view: most sites should leave the setting alone. If you do want to know what AI results are worth to you, test the switch on one section of your site first, not the whole thing. Search Console lets you do that.

What the test found

Shepard published the experiment on his newsletter, Zyppy Signal. The setup:

  • Sept. 1, 2026: he excluded the site using Search Console's Search generative AI control. AI feature impressions fell by more than 99%. He says the site stopped showing in AI results within about an hour.
  • Sept. 29: he opted back in, and impressions returned almost as fast.
  • He compared total clicks in the 28 days without AI features with the 28 days after, then checked the result against matched queries and ranking positions to limit the effect of ranking and query changes.

The numbers he reports:

  • Before the switch, AI Overviews and AI Mode showed the site 500 to 1,000 times a day, about 10% of its Google impressions.
  • Clicks from AI results ran at an estimated 0.9% click-through rate (he gives a range of 0.7% to 1.1%), against about 0.4% for the site overall.
  • That works out to roughly 21% of search clicks coming from AI features. Adding a small lift on the site's regular listings while it was in AI Overviews (about 2.3% higher click-through rate), he puts the total at about 23%.

He is careful about what this proves. It is one site, and one that has done a lot of work to get cited by AI, which he says "may be why so many of our clicks came from AI search." He calls the result a "directional yardstick." He doesn't publish raw click counts, so we can't check the math ourselves.

Does opting out of AI Overviews affect rankings?

Google says no. Its help page says the control "isn't used as a ranking or inclusion signal affecting other parts of Search." Your regular blue links stay where they are.

That cuts both ways. Opting out costs you nothing in regular rankings, but it gains you nothing there either. The only thing it changes is whether you appear in AI answers, and Shepard's test suggests those answers can send more clicks per impression than regular results do.

What the setting actually covers

The control sits in Search Console under Settings > Search generative AI. Google says it was rolled out to all websites worldwide as of Aug. 31, 2026. Per Google's help page:

  • What it removes: your links and content from AI Overviews, AI Mode and generative AI features in Google Discover. Your content also stops being used as input for those answers.
  • What it doesn't touch: AI training (that is Google-Extended in robots.txt), your presence in regular Search (that is noindex), or settings in Merchant Center and Google Ads.
  • How fast: Google says content should be excluded "within 1-2 days," and some may take longer. Shepard saw it happen within about an hour.
  • Who it applies to: the property you set it on and everything under it. Child properties inherit the setting unless their owner changes it.

How much traffic comes from AI Overviews on your site?

Search Console can't tell you directly. Its generative AI performance report shows impressions in AI Overviews and AI Mode, by page, country, device and date, but no clicks. That gap is why Shepard had to switch the feature off to measure it.

You can get a rough sense without opting out. Open the generative AI report and the regular Performance report for the same dates, and compare AI impressions with total impressions. On Shepard's site AI was about a tenth of impressions and, at roughly twice the click-through rate, about a fifth of clicks. Your ratio could be very different, so treat that only as a reason to test, not as an answer.

How to test it on one section first

Shepard's design compares one period with another, so anything else that changed between September and October (seasonality, a ranking shift, a news cycle) lands in the result. The inheritance rule in Search Console allows a cleaner test: switch off one part of the site and keep the rest as a control, over the same dates.

  1. Pick a section with steady traffic and a meaningful number of AI impressions in the generative AI report, such as /blog/ or /guides/.
  2. Add a URL-prefix property for that folder in Search Console (for example https://example.com/blog/), if you don't have one. It inherits the site's setting by default.
  3. Set that property to exclude its content from Search generative AI features. Leave the rest of the site included.
  4. Run it for 28 days, then switch the section back on.
  5. Compare the change, not the totals. For the test section and the rest of the site, take clicks in the test period divided by clicks in the 28 days before. If the test section fell 15% while the rest of the site was flat, AI features were worth roughly 15% of that section's clicks. If both moved together, the setting wasn't what moved them.

Pick a section you can afford to lose some clicks on for a month. That is the cost of the answer.

Who might still opt out

We can see a case for it in a few places: a site whose content is fully answered in the AI summary (definitions, quick facts, conversions), or a publisher making a point about how its content is used. For a business site that wants calls, bookings or sales, we think the risk runs the other way. Being left out of the answer means a competitor is cited instead.

For this site, we are staying in. A news site wants to be the source an AI answer links to, the same reason we allow AI search crawlers.

What we don't know yet

  • How typical the 21% is. It comes from one site with a strong AI presence. Shepard says other sites could be higher or lower, and no one has published a second test yet.
  • Whether Google will ever report AI clicks. The generative AI report counts impressions only.
  • Whether switching back is always fast. Google's help page doesn't say how long re-inclusion takes. On Shepard's site it was quick, but that is one data point.